Architecture/core UI separation by HarshK97 · Pull Request #1 · HarshK97/diffmantic.nvim

HarshK97 · 2026-02-14T05:57:36Z

No description provided.

- added enrich_update_actions_with_semantic() to analyze update actions - detect leaf changes within updated nodes using semantic.find_leaf_change() - extracting rename pairs for identifier changes

- added explicit rename action generation from update semantic leaf changes - added metadata on rename actions - added reusable range metadata helper, can be used for other actions to expose it to ui

gives around 100 renames in 2 ms so pretty fast

- normalizing each actions with build_action function - which carries type, kind, src_node, dst_node, src_range, dst_range, and other temporary compatibility fields: nodes, target so the engine dont explode right now

- now gives internal summaries of actions - this includes, summary.counts, summary.moves, summary.renames, summary.inserts, summary.deletes, summary.updates - from_line and to_line, is include in move summary - now core.generate_actions(...) returns actions, timings, summary

instead of hard coding the language node types will use query files to get the node type and use them (learned the hard way)

using the query files now to detect the actions, but now everything which is not mapped together is shown inserted/deleted will need to fix that... somehow

i was having the highlight based on both if it was mapped and if it had any descendant which was mapped to making everything insert/delete

added a stable ordering for nodes at each height level in top down matching. now nodes are sorted by their source position(start row/col, end row/col) and then by type, this prevents non-deterministic matching regardless of tree traversal order

… recovery passes similar to top-down ordering position deterministically, providing consistent order based on their source position. Bottom up pass sorts nodes by heights then position where recovery only sorts by position only

replaced the broad identifier rename captures with declaration focused captures across different languages, this tightens semantic rename check for field nodes to only accept name/key positions, this lets us reduce noisy rename detections from normal identifier usages and improves rename action quality

added a new core/payload.lua to build spans/signs/virt_text from normalized actions wired actions.lua to use payload.lua this keeps ui focused on consuming the payload instead of inferring semantic diffs itself

removed semantic/highlight diff logic from ui/renderer.lua and now renderer uses the core payload to show the highlights added ui/signs.lua for sign style/glyph/priority handling cleaned ui/helpers.lua by removing old semantic/internal diff helpers no longer needed in ui

renamed internal diff highlight group from Diff*Text/Sign to Diffmantic*

- added src_fillers/dst_fillers to payload output structure - generating filler entries for deletions/insertion/movement - handle trailing blank lines and top level nodes - applied fillers as virtual text as the native filler lines seem to be a native diff feature

…ysis.lua Normalized action objects around src/dst/metadata so actions.lua becomes core payload source. Replaced the update semantic storage from action.semantic to action.analysis

… move/rename operations - Add range_contains() to check if one range is within another - Add collect_suppressed_rename_pairs() to track renames within declarations and suppress redundant annotations - Add visual text annotations for move operations showing source/destination line numbers with arrows - Add visual text annotations for renames showing old/new names - Remove unused extra_src_fillers and extra_dst_fillers handling - Clean up trailing whitespace

Copilot

Pull request overview

This PR advances the “core vs UI” separation by moving semantic/rename detection and diff analysis into core modules, introducing Treesitter role queries across multiple languages, and simplifying the UI renderer to consume enriched action metadata (including rename + filler rendering).

Changes:

Add Treesitter role indexing + language queries to classify semantic constructs (functions/classes/variables/assignments/imports/returns/rename identifiers).
Extend action generation to emit rename actions and enrich update actions with analyzed hunks/timings, and make node iteration deterministic.
Refactor UI rendering into smaller modules (signs + filler) and update benchmark/init to pass buffer context into core.

Reviewed changes

Copilot reviewed 22 out of 23 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
test/benchmark.lua	Adds rename cases and passes `src_buf`/`dst_buf` into action generation; reports rename timings/actions.
queries/typescript/diffmantic.scm	Adds TS query captures for semantic roles + rename identifiers.
queries/python/diffmantic.scm	Adds Python query captures for semantic roles + rename identifiers.
queries/lua/diffmantic.scm	Adds Lua query captures for semantic roles + rename identifiers.
queries/javascript/diffmantic.scm	Adds JS query captures for semantic roles + rename identifiers.
queries/go/diffmantic.scm	Adds Go query captures for semantic roles + rename identifiers.
queries/cpp/diffmantic.scm	Adds C++ query captures for semantic roles + rename identifiers.
queries/c/diffmantic.scm	Adds C query captures for semantic roles + rename identifiers.
queries/fallback.scm	Adds a fallback query used when language-specific queries are unavailable.
lua/diffmantic/ui/signs.lua	New sign rendering helper with style + priority handling.
lua/diffmantic/ui/filler.lua	New filler-line computation/application to align moved/inserted/deleted blocks.
lua/diffmantic/ui/renderer.lua	Refactors renderer to consume enriched action ranges/hunks and delegate to signs/filler.
lua/diffmantic/ui/helpers.lua	Removes prior diff logic, leaving only inline virt-text helper.
lua/diffmantic/init.lua	Updates highlight setup (with ColorScheme autocmd) and passes buffers into action generation.
lua/diffmantic/core/top_down.lua	Adds stable ordering for node processing at each height.
lua/diffmantic/core/recovery.lua	Makes recovery application deterministic by sorting `src_info` traversal.
lua/diffmantic/core/bottom_up.lua	Integrates role index for name-based matching and deterministic traversal.
lua/diffmantic/core/semantic.lua	Introduces core semantic helpers (leaf diffs, rename-identifier heuristics, fragments).
lua/diffmantic/core/roles.lua	New module to build/query a capture index from `diffmantic` Treesitter queries.
lua/diffmantic/core/analysis.lua	New module to compute update hunks and suppress rename-noise within updates.
lua/diffmantic/core/actions.lua	Major refactor: richer action payloads, rename emission, timings, and update enrichment.
lua/diffmantic/core/payload.lua	New (currently unreferenced) module containing render-enrichment logic.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-14T06:07:43Z

lua/diffmantic/core/roles.lua

+	for _, captured in pairs(by_capture) do
+		if node:equal(captured) or node:child_with_descendant(captured) then
+			return captured
+		end
+	end
+
+	return nil


find_descendant_with_capture iterates pairs(by_capture), which is non-deterministic and can return a nested capture (e.g., a nested function name) instead of the closest/outer one. This can make name extraction flaky and break matching/rename detection. Consider selecting the best candidate deterministically (e.g., smallest range / earliest in document order) or iterating captures in-order for the given subtree rather than using pairs over a map.

Suggested change

for _, captured in pairs(by_capture) do

if node:equal(captured) or node:child_with_descendant(captured) then

return captured

end

end

return nil

local best = nil

local best_len_rows, best_len_cols = nil, nil

local best_start_row, best_start_col = nil, nil

for _, captured in pairs(by_capture) do

if node:equal(captured) or node:child_with_descendant(captured) then

local start_row, start_col, end_row, end_col = captured:range()

local len_rows = end_row - start_row

local len_cols = end_col - start_col

if not best then

best = captured

best_len_rows, best_len_cols = len_rows, len_cols

best_start_row, best_start_col = start_row, start_col

else

-- Prefer the smallest range; if equal, prefer earlier in document order.

if len_rows < best_len_rows

or (len_rows == best_len_rows and (len_cols < best_len_cols

or (len_cols == best_len_cols and (start_row < best_start_row

or (start_row == best_start_row and start_col < best_start_col)))))) then

best = captured

best_len_rows, best_len_cols = len_rows, len_cols

best_start_row, best_start_col = start_row, start_col

end

end

end

end

return best

Copilot · 2026-02-14T06:07:44Z

lua/diffmantic/core/semantic.lua

+	if math.abs(#src_leaves - #dst_leaves) > 2 then
+		return {}
+	end
+


In find_leaf_changes, the math.abs(#src_leaves - #dst_leaves) > 2 guard is unreachable because the previous check returns immediately when the lengths differ. If the intent is to allow small leaf-count differences, the first condition should be relaxed; otherwise the second check should be removed to avoid dead code and confusion.

Suggested change

if math.abs(#src_leaves - #dst_leaves) > 2 then

return {}

end

Copilot · 2026-02-14T06:07:44Z

lua/diffmantic/core/actions.lua

+	local function has_kind(node, index, kind)
+		return index and roles.has_kind(node, index, kind) or false
+	end
+
+	local function is_significant(info, index)
+		local node = info.node
+		if has_kind(node, index, "function")
+			or has_kind(node, index, "class")
+			or has_kind(node, index, "variable")
+			or has_kind(node, index, "assignment")
+			or has_kind(node, index, "import")
+			or has_kind(node, index, "return")
+			or has_kind(node, index, "preproc")
+		then
+			return true
+		end
+		return significant_types[info.type] or false
+	end
+
+	local function is_transparent_update_ancestor(info, index)
+		if has_kind(info.node, index, "class") then
+			return true
+		end
+		return transparent_update_ancestors[info.type] or false
+	end
+
+	local function is_movable(info, index)
+		if has_kind(info.node, index, "function") or has_kind(info.node, index, "class") then
+			return true
+		end
+		return movable_types[info.type] or false
+	end


roles.has_kind(..., "function"/"class"/"variable"/...) returns true for any node captured as outer/name/body (per CAPTURE_BY_KIND). Using it in is_significant / is_movable can accidentally mark name identifiers or bodies as significant/movable, which may cause noisy insert/delete/update actions for sub-nodes when mappings are incomplete. Consider making the kind checks refer only to the *.outer captures (or adding a separate has_outer_kind helper) so only the declaration/container node is treated as significant.

Copilot · 2026-02-14T06:07:44Z

lua/diffmantic/ui/filler.lua

+			if src_count > 0 then
+				local src_block_end = sec > 0 and (ser + 1) or ser
+				local best_dst_row = nil
+				local best_src_row = math.huge
+				for _, other in ipairs(actions) do
+					if other ~= action and other.src_node and other.dst_node then
+						local osr = select(1, other.src_node:range())
+						local odr = select(1, other.dst_node:range())
+						if osr >= src_block_end and osr < best_src_row then
+							best_src_row = osr
+							best_dst_row = odr
+						end
+					end
+				end


compute() does an O(n) scan over actions for every move action to find best_dst_row (nested loop), which becomes O(n²) in large diffs and can dominate runtime (benchmarks go up to thousands of nodes). Consider precomputing the “next move after this source block” mapping once (e.g., sort moves by src_row and keep a parallel list of dst_row) so each move can find its anchor in O(log n) or O(1).

Copilot · 2026-02-14T06:07:45Z

lua/diffmantic/core/payload.lua

+local semantic = require("diffmantic.core.semantic")
+
+local M = {}
+local RENDER_UPDATE_TOKENS = true
+
+local function node_range(node)
+	if not node then
+		return nil
+	end
+	local ok, sr, sc, er, ec = pcall(function()


lua/diffmantic/core/payload.lua is added but doesn’t appear to be referenced anywhere in the repository (no require("diffmantic.core.payload") usages). If it’s not part of the current architecture split, consider removing it from this PR to avoid carrying a large unused module; otherwise, wire it into the code path or add a brief comment explaining its intended entry point.

Copilot · 2026-02-14T06:07:45Z

lua/diffmantic/ui/helpers.lua

@@ -236,278 +13,4 @@ function M.set_inline_virt_text(buf, ns, row, col, text, hl)
 	pcall(vim.api.nvim_buf_set_extmark, buf, ns, row, col, opts)
 end


lua/diffmantic/ui/helpers.lua now only contains set_inline_virt_text, but there are no remaining references to diffmantic.ui.helpers or set_inline_virt_text in the codebase. Consider removing this module (or reusing it from renderer.lua) to avoid dead code.

- action range construction now uses normalized src/dst fields - summary tables read metadata/src/dst directly inside of legacy code

Simplify parameter identifier extraction and suppress noisy field declaration renames that lack semantic context. - Change walk() to first_identifier() - extract only first identifier per parameter instead of recursively collecting all - Remove type_identifier from identifier matching - Remove Pass 1b strict fallback seed collection logic - Disable rename detection for field_declaration declarators - Update C/C++ queries to exclude field_identifier from init_declarator

prevents noisy rename detection for identifiers within class bodies that are not the class name itself suppressing update actions for class bodies in favour of move/insert/delete - added is_class_like_context check to exclude non-name identifiers in class bodies from rename detection - added should_emit_update to suppress update for class/struct/types bodies

…count tracking - added range_text() helper to extract text from buffer ranges - added action_pair_key() to generate unique keys for action pairs - removed base_count field from filler entries (unused) - removed move_anchor_after_blank_run() in favor of more precise count_filler_lines_at() calculation

…me heuristics

…ursion

…lignment

…rendering

…ght tuning

… update

…tching

…acement scaffolding

…covered insert/delete fillers

…s (skip action-level duplicates)

…hanged spans

…move false core diff noise

…uffers

Copilot

Pull request overview

Copilot reviewed 38 out of 39 changed files in this pull request and generated 11 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-24T17:01:40Z

test/comparison/after.lua

+		name = username,
+		email = email,
+		active = true,
+		created_at = nil,


Missing role field assignment in the Lua create_user function. The after.lua version (line 28-34) creates a user table without setting the role field, even though the role parameter is passed in and available. This differs from other language implementations where the role is properly assigned to the user object.

Suggested change

created_at = nil,

created_at = nil,

role = role,

Copilot · 2026-02-24T17:01:40Z

test/comparison/after.c

+struct User create_user(const char *username, const char *email, const char *role) {
+    if (!validate_email_address(email)) {
+        fprintf(stderr, "Invalid email format\n");
+    }
+
+    struct User user;
+    memset(&user, 0, sizeof(user));
+    strncpy(user.name, username, sizeof(user.name) - 1);
+    strncpy(user.email, email, sizeof(user.email) - 1);
+    user.active = 1;
+    user.created_at = NULL;
+    (void)role;
+    return user;


The role parameter is accepted but not used in the create_user function. Similar to the C++ version, the function accepts a role parameter (line 29) but explicitly marks it as unused with (void)role on line 40. The User struct (line 7-12) also doesn't have a role field. This differs from before.c where the role was properly assigned using strncpy.

Copilot · 2026-02-24T17:01:41Z

lua/diffmantic/ui/filler.lua

+	end
+	if inner.end_row == outer.end_row and inner.end_col > outer.end_col then
+		return false
+	end


Inconsistent indentation: line 50 uses a tab character while surrounding lines use tabs. This specific line appears to have unusual indentation that may cause visual inconsistency.

Copilot · 2026-02-24T17:01:41Z

test/comparison/after.lua

+	local x = 10
+	return username.name .. " <" .. username.email .. ">" .. x
+end
+


The format_user_display function has unexpected logic. A local variable x is defined as 10 and then concatenated to the user display string. This produces output like "John Doe <john@example.com>10" which differs from all other language implementations that just return "name <email>". This appears to be test data rather than production code, but the logic is inconsistent with the function's apparent purpose.

Suggested change

local x = 10

return username.name .. " <" .. username.email .. ">" .. x

end

return username.name .. " <" .. username.email .. ">"

end

Copilot · 2026-02-24T17:01:41Z

test/comparison/after.ts

+function getUserPermissions(role: Role): string[] {
+  if (role === "member") {
+    return ["read"];
+  }
+  if (role === "editor") {
+    return ["read", "write"];
+  }
+  if (role === "admin") {
+    return ["read", "write", "delete", "manage"];
+  }
+  return ["read", "write", "delete", "manage", "configure"];
+}


Missing handling of "editor" role in getUserPermissions. The function checks for "member" role (line 46) but the Role type includes "editor" and other valid roles. When role is "editor", the function will fall through to line 55 and return the superadmin permissions array, which is incorrect. An explicit check for "editor" role should be added to return ["read", "write"].

Copilot · 2026-02-24T17:01:42Z

test/comparison/after.cpp

+User create_user(const std::string &username, const std::string &email, const std::string &role = ROLE) {
+    if (!validate_email_address(email)) {
+        throw std::runtime_error("Invalid email format");
+    }
+
+    User user;
+    user.name = username;
+    user.email = email;
+    user.active = true;
+    user.created_at = "";
+    (void)role;
+    return user;


The role parameter is accepted but not used in the create_user function. The function accepts a role parameter with a default value (line 30), but the User struct created on line 35 doesn't include a role field assignment. The role parameter is cast to void on line 40, which explicitly marks it as unused. This differs from the pattern in before.cpp where role was properly used.

Copilot · 2026-02-24T17:01:42Z

test/comparison/after.cpp

+struct User {
+    std::string name;
+    std::string email;
+    bool active;
+    std::string created_at;
+};


The User struct is missing the role field. In before.cpp (line 18-23), the User struct has a role field, but in after.cpp (line 8-13) it's been removed. However, the create_user function still accepts a role parameter. This is a breaking change to the data structure that should be addressed.

Copilot · 2026-02-24T17:01:42Z

lua/diffmantic/core/analysis.lua

+		while i <= len do
+			local ch = text:sub(i, i)
+			if ch:match("%s") then
+				i = i + 1
+			elseif ch:match("[%w_]") then
+				local j = i + 1
+				while j <= len and text:sub(j, j):match("[%w_]") do
+					j = j + 1
+				end
+				table.insert(tokens, { text = text:sub(i, j - 1), start_col = i, end_col = j - 1 })
+				i = j
+			else
+				-- Keep punctuation/granular symbols as single-char tokens so we can
+				-- align shared delimiters (e.g. closing quote) and avoid off-by-one spans.
+				table.insert(tokens, { text = ch, start_col = i, end_col = i })
+				i = i + 1
+			end
+		end
+		return tokens
+	end


Inconsistent indentation: lines 50-68 use inconsistent indentation (tabs vs spaces). The function body starting at line 50 should maintain consistent indentation with the rest of the file.

Suggested change

while i <= len do

local ch = text:sub(i, i)

if ch:match("%s") then

i = i + 1

elseif ch:match("[%w_]") then

local j = i + 1

while j <= len and text:sub(j, j):match("[%w_]") do

j = j + 1

end

table.insert(tokens, { text = text:sub(i, j - 1), start_col = i, end_col = j - 1 })

i = j

else

-- Keep punctuation/granular symbols as single-char tokens so we can

-- align shared delimiters (e.g. closing quote) and avoid off-by-one spans.

table.insert(tokens, { text = ch, start_col = i, end_col = i })

i = i + 1

end

end

return tokens

end

while i <= len do

local ch = text:sub(i, i)

if ch:match("%s") then

i = i + 1

elseif ch:match("[%w_]") then

local j = i + 1

while j <= len and text:sub(j, j):match("[%w_]") do

j = j + 1

end

table.insert(tokens, { text = text:sub(i, j - 1), start_col = i, end_col = j - 1 })

i = j

else

-- Keep punctuation/granular symbols as single-char tokens so we can

-- align shared delimiters (e.g. closing quote) and avoid off-by-one spans.

table.insert(tokens, { text = ch, start_col = i, end_col = i })

i = i + 1

end

end

return tokens

end

Copilot · 2026-02-24T17:01:43Z

lua/diffmantic/init.lua

 		vim.cmd("tabnew")
 		vim.cmd("edit " .. file1)
 		buf1 = vim.api.nvim_get_current_buf()
-		local win1 = vim.api.nvim_get_current_win()
+		win1 = vim.api.nvim_get_current_win()

 		vim.cmd("vsplit " .. file2)
 		buf2 = vim.api.nvim_get_current_buf()


vim.cmd is being built by concatenating untrusted path arguments (file1/file2) directly into an Ex command string, which allows a crafted filename or argument containing | or newlines to inject additional Vim commands (command injection). An attacker controlling the file path (e.g. via a repository containing a file named foo|:!rm -rf ~ or via a crafted :Diffmantic invocation) could cause arbitrary Ex commands to run in the user’s editor. Use the structured Lua API for commands (e.g. passing the path via an args/mods table or equivalent) or robustly escape/quote the path so that user-controlled strings cannot break out of the edit/vsplit command.

Copilot · 2026-02-24T17:01:43Z

lua/diffmantic/init.lua

 		local expanded_path = vim.fn.expand(file1)

 		vim.cmd("vsplit " .. expanded_path)
 		buf2 = vim.api.nvim_get_current_buf()
-		local win2 = vim.api.nvim_get_current_win()
-
-		vim.wo[win1].scrollbind = true
-		vim.wo[win1].cursorbind = true
-		vim.wo[win2].scrollbind = true
-		vim.wo[win2].cursorbind = true
+		win2 = vim.api.nvim_get_current_win()


vim.cmd is called with expanded_path interpolated directly into the command string, so a filename that expands to a value containing | or newlines can inject extra Ex commands (command injection) when users run :Diffmantic. Because expanded_path is derived from vim.fn.expand(file1), a maliciously named file combined with a trusted mapping or autocmd that calls this function could trigger arbitrary command execution in Neovim. Switch to the structured command API (passing the path in a dedicated args field) or escape/quote expanded_path before concatenation so that special command separators cannot be interpreted.

HarshK97 added 19 commits February 8, 2026 08:31

moved helpers from ui to core for separating core and ui

1bc1b2b

feat: enriching update actions with semantic info

733be00

- added enrich_update_actions_with_semantic() to analyze update actions - detect leaf changes within updated nodes using semantic.find_leaf_change() - extracting rename pairs for identifier changes

feat(core): added explicit rename action with metadata

f30a728

- added explicit rename action generation from update semantic leaf changes - added metadata on rename actions - added reusable range metadata helper, can be used for other actions to expose it to ui

benchmark: updated the benchmark.lua to show the rename action time

e97ddb9

gives around 100 renames in 2 ms so pretty fast

feat(core): normalizing internal actions

e07fd8a

- normalizing each actions with build_action function - which carries type, kind, src_node, dst_node, src_range, dst_range, and other temporary compatibility fields: nodes, target so the engine dont explode right now

feat(core): added query files to support languages via tree-sitter

c2eea43

instead of hard coding the language node types will use query files to get the node type and use them (learned the hard way)

feat(core): using query roles for semantic and action classification

db69c7f

using the query files now to detect the actions, but now everything which is not mapped together is shown inserted/deleted will need to fix that... somehow

fix: highlighting anything not tagged as insert/delete

7330f96

i was having the highlight based on both if it was mapped and if it had any descendant which was mapped to making everything insert/delete

feat(core): moved render payload construction to core engine

e244f21

added a new core/payload.lua to build spans/signs/virt_text from normalized actions wired actions.lua to use payload.lua this keeps ui focused on consuming the payload instead of inferring semantic diffs itself

refactor(ui): updated namespace highlights groups to Diffmantic*

4aa9914

renamed internal diff highlight group from Diff*Text/Sign to Diffmantic*

updated the diff viewer image

ee3b7b2

refactor(core): normalized action payload and move enrichment to anal…

a459f44

…ysis.lua Normalized action objects around src/dst/metadata so actions.lua becomes core payload source. Replaced the update semantic storage from action.semantic to action.analysis

HarshK97 requested a review from Copilot February 14, 2026 05:57

Copilot started reviewing on behalf of HarshK97 February 14, 2026 05:58 View session

Copilot AI mentioned this pull request Feb 14, 2026

[WIP] Update architecture for core UI separation #2

Closed

Repository owner deleted a comment from Copilot AI Feb 14, 2026

This comment was marked as spam.

Sign in to view

Copilot AI reviewed Feb 14, 2026

View reviewed changes

refactor(core): normalized action src/dst + metadata summary

e554a4d

- action range construction now uses normalized src/dst fields - summary tables read metadata/src/dst directly inside of legacy code

Copilot AI mentioned this pull request Feb 14, 2026

[WIP] Implement core and UI separation in architecture #3

Closed

Repository owner deleted a comment from Copilot AI Feb 14, 2026

delete(core): removed unused core/payload.lua and ui/helpers.lua

f5ada37

HarshK97 force-pushed the architecture/core-ui-separation branch from e3d1ad3 to f5ada37 Compare February 14, 2026 07:05

HarshK97 added 5 commits February 15, 2026 17:37

refactor(ui): removed filler cause it sucks right now

9603a56

feat(queries): add structural captures for C/C++/Go/JS/TS significance

e0a9a04

HarshK97 force-pushed the architecture/core-ui-separation branch 2 times, most recently from 9442d7b to e0a9a04 Compare February 17, 2026 13:34

HarshK97 added 17 commits February 17, 2026 19:08

refactor(core): enforce parent-safe top-down behavior and expand rena…

4b24d8a

…me heuristics

refactor(core): deterministic left-biased recovery with immediate rec…

5c331a1

…ursion

refactor(core): token-level asymmetric hunk classification and span a…

98af7a2

…lignment

refactor(core): key-aware literal overrides and temp-result refactor …

1a9841d

…rendering

refactor(ui): emphasize update-over-move with accent layer and highli…

7a35cca

…ght tuning

test(comparison): add cross-language before/after fixtures and sample…

f1376de

… update

perf(core): optimize preprocess hashing and parent-bucket top-down ma…

21be528

…tching

perf(core): cache role/name resolution and speed bottom-up matching

a942f32

perf(core): switch recovery to sparse worklist with bounded LCS

0b653e9

perf(core): reuse role indexes and memoize action precompute ancestry

47bb9c5

test(perf): benchmark timing breakdown

988bad2

feat(ui/filler): introduce semantic filler pipeline and move-aware pl…

3c9bac3

…acement scaffolding

fix(ui/filler): ignore render_as_change update hunks when deriving un…

3caeced

…covered insert/delete fillers

fix(ui/renderer): render insert/delete as change only via update hunk…

7657c42

…s (skip action-level duplicates)

refactor(core/analysis): refine single-line change hunks to minimal c…

219f92a

…hanged spans

fix(core/analysis): collapse identical insert/delete hunk pairs to re…

96e4e3b

…move false core diff noise

feat(ui): pass mappings to renderer and disable diagnostics in diff b…

74e1038

…uffers

HarshK97 requested a review from Copilot February 24, 2026 16:55

Copilot started reviewing on behalf of HarshK97 February 24, 2026 16:55 View session

Copilot AI reviewed Feb 24, 2026

View reviewed changes

HarshK97 marked this pull request as ready for review February 25, 2026 11:32

HarshK97 merged commit 5fd0122 into main Feb 25, 2026
4 checks passed

HarshK97 deleted the architecture/core-ui-separation branch February 28, 2026 10:36

-	for _, captured in pairs(by_capture) do
-		if node:equal(captured) or node:child_with_descendant(captured) then
-			return captured
-		end
-	end
-	return nil
+	local best = nil
+	local best_len_rows, best_len_cols = nil, nil
+	local best_start_row, best_start_col = nil, nil
+	for _, captured in pairs(by_capture) do
+		if node:equal(captured) or node:child_with_descendant(captured) then
+			local start_row, start_col, end_row, end_col = captured:range()
+			local len_rows = end_row - start_row
+			local len_cols = end_col - start_col
+			if not best then
+				best = captured
+				best_len_rows, best_len_cols = len_rows, len_cols
+				best_start_row, best_start_col = start_row, start_col
+			else
+				-- Prefer the smallest range; if equal, prefer earlier in document order.
+				if len_rows < best_len_rows
+					or (len_rows == best_len_rows and (len_cols < best_len_cols
+						or (len_cols == best_len_cols and (start_row < best_start_row
+							or (start_row == best_start_row and start_col < best_start_col)))))) then
+					best = captured
+					best_len_rows, best_len_cols = len_rows, len_cols
+					best_start_row, best_start_col = start_row, start_col
+				end
+			end
+		end
+	end
+	return best

		@@ -236,278 +13,4 @@ function M.set_inline_virt_text(buf, ns, row, col, text, hl)
		pcall(vim.api.nvim_buf_set_extmark, buf, ns, row, col, opts)
		end

Conversation

HarshK97 commented Feb 14, 2026

Uh oh!

This comment was marked as spam.

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 14, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants